Measuring rates of brain atrophy from serial magnetic resonance imaging (MRI) scans is gaining acceptance as a valid noninvasive biomarker of disease progression in neurodegenerative conditions. To date, the two most widely used measurement techniques are, manual segmentation of specific structures (e.g. hippocampus), or semi-automated image registration and subtraction to generate whole brain atrophy rates. Both of these techniques, while useful, have important limitations in neurodegenerative conditions that are characterized by atrophy of select regions of the brain. A prototypical neurodegenerative condition that is characterized by selective atrophy is frontal temporal dementia (FTD). A class of computational techniques that we refer to in this application as tensor based morphometry (TBM) seem ideally suited to measure atrophy rates of defined regions/lobes of the brain from serial MRI scans. We have developed a prototype version of a TBM algorithm and the associated software code, but have not yet performed the clinical comparison studies necessary for validation. Another largely ignored but important technical topic addressed in this application is rigorous evaluation of sources of and methods to correct geometric distortion in MR images. This project has two major goals. The first is to optimize our prototype TBM method and to compare it to other approaches for validation purposes. We will also evaluate methods for correcting geometric distortion in MRI. This validation and testing work is the objective of Aims 1 and 2. The second major goal is to prospectively test the hypothesis that the regional atrophy rate measurement method developed during the technical phase of the project will provide group specific signatures that are characteristic of particular clinical phenotypes, specifically normal aging, AD, and FTD. We will test the hypothesis that TBM regional atrophy rate measures show greater clinical group specificity that global atrophy rate measures. We will also include pathological verification and correlation when possible.